PruneV is a tool which reduces the injection space of the SEU injection simulation tool. This is done specifically for injections carried out on the register file of the processor core.
PruneV works by parsing an instruction trace of a firmware running on a given core. An example of the Coremark firmware running on an Ibex core is shown below.
| Time | Cycle | PC | Insn | Decoded instruction | Register and memory contents |
|---|---|---|---|---|---|
| 80000 | 6 | 80000080 | 0040006f | jal x0,80000084 | x0=0x00000000 |
| 90000 | 7 | 80000084 | 02040117 | auipc x2,0x2040 | x2=0x82040084 |
| 100000 | 8 | 80000088 | f7c10113 | addi x2,x2,-132 | x2:0x82040084 x2=0x82040000 |
| 110000 | 9 | 8000008c | 810002b7 | lui x5,0x81000 | x5=0x81000000 |
| 120000 | 10 | 80000090 | 00028293 | addi x5,x5,0 | x5:0x81000000 x5=0x81000000 |
| 130000 | 11 | 80000094 | 30529073 | csrrw x0,mtvec,x5 | x5:0x81000000 x0=0x00000000 |
| 140000 | 12 | 80000098 | 08028293 | addi x5,x5,128 | x5:0x81000000 x5=0x81000080 |
| 160000 | 14 | 8000009c | 00028067 | jalr x0,0(x5) | x5:0x81000080 x0=0x00000000 |
| 180000 | 16 | 81000080 | 0100006f | jal x0,81000090 | x0=0x00000000 |
Some of these instructions are standard RISC-V instructions, which are supported by PruneV, and others are specific to the Ibex core, which are not supported as of now.
The working principles of PruneV is two-fold. The first is that a register is not sensitive between the cycles of one instruction, and another, if the latter instruction completely overrides the values of that register. An example of such an instruction is a load instruction which takes a value from memory, and stores it in the register. The second principle is that if a register is read from at cycle 
To use PruneV the user must define the custom_line_parser function in the main.py script. This is only necessary if the user wants to parse lines not parsable by the custom line_decoder (which the user should create in the line_and_decoders folder, by following the pre-existing decoders).
The user should choose an instruction library from the InstructionLibrary enum type (new libraries should be added to this enum), and registers to exclude from the pruning. This is showcased in the main.py script as well.
The result object created by the CoreTraceParser.parse() method is a matrix, with the cycles of the instruction trace represented by the rows, and the registers represented by the columns. A 0-entry in this matrix indicates that the register is not sensitive at that given cycle, and a series of identical non-zero values indicate that a fault-injection on any of these cycles, for that register, will result in the same outcome for the simulation.
Below is a flowchart, which should give an idea of how PruneV works.
